Study design and hypothesis testing: issues in the evaluation of evidence from research in nutritional epidemiology1,2 Jo L Freudenheim ABSTRACT In addressing questions about the relations of individuals under study (2, 11, 12). The classic epidemiologic dietary factors to disease in human populations, epidemiologic study designs, namely ecologic, case-control, cohort, and clini- studies must account for the complexity of dietary habits, the cal trials, have been used to examine the relations of dietary intercorrelations among dietary habits, and the correlations of those practices to health and disease. These study designs and their Downloaded from habits with other behaviors. Furthermore, for studies of chronic strengths and weaknesses are described briefly in this article. A disease, relevant dietary exposures may occur over decades. The more complete, general discussion of these study designs can be classic epidemiologic study designs have been used to examine the found elsewhere (13, 14). An understanding of the strengths and associations between diet and disease; the strengths and weaknesses weaknesses of epidemiologic studies, and of studies about diet in of those designs must be considered. Concerns have been raised particular, is essential for their interpretation for the purposes of regarding the validity of the measures of diet, the differential recall setting public policy. www.ajcn.org of diet by diseased individuals in case-control studies, and confounding by other related factors in both case-control and cohort studies. In clinical trials there may be difficulties in effecting the EPIDEMIOLOGIC STUDY DESIGNS necessary dietary changes, especially for macronutrients, and there The study designs used in nutritional epidemiology include are also concerns about those circumstances in which participants ecologic studies, which investigate diet and disease at the popu- at Harvard Libraries on February 9, 2008 cannot be blinded to their treatment. For case-control and cohort lation level, and several study designs (cross-sectional, case-con- studies and for some clinical trials, intercorrelations among trol, cohort, and clinical trial) that address such questions at the nutrients are a concern in the identification of factors that are level of the individual. In ecologic studies the unit of study is a important in the etiology of disease. It is important to understand population, which is usually defined by geography; in each pop- these considerations when interpreting nutritional epidemiologic ulation a measure of disease frequency is correlated with a meas- studies for the purpose of setting public policy. No one study can be ure of nutrient exposure. Such correlations are particularly use- considered definitive in the understanding of a diet–disease relation. ful in the generation of hypotheses about dietary factors that may However, epidemiologic findings from multiple studies taken be associated with variations in disease rates. Ecologic studies together can contribute significantly to our understanding of diet in often have limitations because the measures of dietary exposure relation to disease in humans. Am J Clin Nutr 1999;69(suppl): are frequently made for other purposes (eg, food disappearance 1315S–21S. data collected for economic purposes) and therefore are not nec- essarily in a form that accurately describes population exposure. KEY WORDS Nutritional epidemiology, chronic disease More importantly, results from ecologic studies are inherently epidemiology, study design, case-control study, cohort study, restricted to an evaluation at the population level and do not pro- clinical trial, disease etiology, diet, nutrition, dietary habits, vide evidence about whether the individuals in the population eating habits, humans who get the disease under study are those who experienced the exposure being measured in the population. Furthermore, in such studies it is difficult to account for other factors correlated with INTRODUCTION the exposure of interest, which may account for variation among Epidemiologic studies address the important but tangled ques- study units in the observed rate of disease. tions concerned with identifying the factors that lead to disease The epidemiologic study designs that address the relations in human populations. In nutritional epidemiology the focus is between dietary exposures and disease risk at the individual level on dietary factors related to disease while taking into account other, nondietary factors. Such studies must account for the com- plexity of dietary habits (1, 2), the intercorrelations among 1 From the Department of Social and Preventive Medicine, University at dietary factors (3–5), and the correlations between dietary habits Buffalo. and other behaviors that have health consequences (6–10). In the 2 Address reprint requests to JL Freudenheim, Department of Social and study of chronic diseases, these analyses may need to evaluate Preventive Medicine, 270 Farber Hall, Buffalo, NY 14214. E-mail: exposures spanning decades or even the entire lifetimes of the [email protected]. Am J Clin Nutr 1999;69(suppl):1315S–21S. Printed in USA. © 1999 American Society for Clinical Nutrition 1315S 1316S FREUDENHEIM are the cross-sectional study, case-control study, cohort study, and jects) from the nondiseased population out of which the case clinical trial. For all of these, there are issues regarding the accu- subjects arose. The control subjects may be selected so that their rate measurement of dietary exposure, which are addressed in distribution on factors such as age and sex reflects the distribu- detail in other articles in this supplement. In all but clinical trials, tion of those factors in the diseased population. there are also issues regarding the extent to which a study is able Case-control studies allow for in-depth inquiry into factors to account for confounding factors (factors correlated with both related to risk of disease, with information collected on an indi- the exposure of interest and the disease outcome). Confounding is vidual basis. These studies are relatively efficient and quick, a concern when the confounding factor is unmeasured, measured with generally lower costs than either cohort studies or clinical with error, or so closely correlated with the dietary exposure of trials. Frequently, case-control studies are population based and interest that the dietary factor and the correlated factor cannot be findings can be generalized, at least to the population under distinguished. Another issue is effect modification, any third fac- study and often to a wider population. Case-control studies can tor that alters the association between exposure and disease. Of focus on dietary exposures in the past, while allowing for the particular interest in this regard is effect modification by genetic issues regarding measurement of diet in the past (2, 11, 12). factors, which may cause interindividual differences in the effect There are several disadvantages to this type of study. One is of an exposure on disease risk. A lack of knowledge about such the difficulty in validating reported exposures, particularly expo- factors may result in an effect of importance in a subgroup being sures in the distant past. Another is the difficulty in identifying masked in that group or, alternatively, an assumption that exposure an appropriate group of control subjects, particularly in studies is related to risk in the entire population when the effect is actually that are not based on geographically defined populations. Ide- limited to a subgroup. An understanding of such variation in ally, once the appropriate control group has been identified, all Downloaded from response to exposures may help explain inconsistencies among the eligible case and control subjects would participate. In fact, this findings of different studies that are related to differences in the is rarely accomplished and there are concerns regarding the rep- distribution of genetic factors in the study samples. Recently resentativeness of those who do participate. Furthermore, there there has been a proliferation of information on biologic variabil- is the concern that because the case subjects are sick they will ity related to genetics. These issues can be addressed by classical think about and report on their diets differently than will the con- epidemiologic methodologies with information regarding the trol subjects, a process called recall bias. www.ajcn.org appropriate factors to be considered and a sufficient sample size. Cohort studies Cross-sectional studies In a cohort study, a group of individuals is identified and their In cross-sectional studies, exposure and disease are assessed exposures to dietary factors and other risk factors of interest are concurrently in individuals selected from a defined population. measured. Assessment of dietary exposures may include both at Harvard Libraries on February 9, 2008 Such studies frequently include a biologic measurement of dis- present and past dietary practices. These individuals are then fol- ease or of nutrient exposure. An example is the work that corre- lowed over time to identify those who develop disease; the meas- lated calcium intake with blood pressure measurements in healthy ured exposures are then used to determine predictors of disease populations (15). Cross-sectional studies provide information risk. The validity of the study is related to the completeness of about disease prevalence and factors associated with that preva- the follow-up.